Class: Rdkafka::Consumer

Inherits:
Object
  • Object
show all
Includes:
Enumerable
Defined in:
lib/rdkafka/consumer.rb,
lib/rdkafka/consumer/headers.rb,
lib/rdkafka/consumer/message.rb,
lib/rdkafka/consumer/partition.rb,
lib/rdkafka/consumer/topic_partition_list.rb

Overview

A consumer of Kafka messages. It uses the high-level consumer approach where the Kafka brokers automatically assign partitions and load balance partitions over consumers that have the same :"group.id" set in their configuration.

To create a consumer set up a Config and call consumer on that. It is mandatory to set :"group.id" in the configuration.

Consumer implements Enumerable, so you can use each to consume messages, or for example each_slice to consume batches of messages.

Defined Under Namespace

Modules: Headers Classes: Message, Partition, TopicPartitionList

Instance Method Summary collapse

Instance Method Details

#assign(list) ⇒ Object

Atomic assignment of partitions to consume

Parameters:

Raises:



162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
 
# File 'lib/rdkafka/consumer.rb', line 162

def assign(list)
  closed_consumer_check(__method__)

  unless list.is_a?(TopicPartitionList)
    raise TypeError.new("list has to be a TopicPartitionList")
  end

  tpl = list.to_native_tpl

  begin
    response = Rdkafka::Bindings.rd_kafka_assign(@native_kafka.inner, tpl)
    if response != 0
      raise Rdkafka::RdkafkaError.new(response, "Error assigning '#{list.to_h}'")
    end
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl)
  end
end

#assignment ⇒ TopicPartitionList

Returns the current partition assignment.

Returns:

Raises:



186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
 
# File 'lib/rdkafka/consumer.rb', line 186

def assignment
  closed_consumer_check(__method__)

  ptr = FFI::MemoryPointer.new(:pointer)
  response = Rdkafka::Bindings.rd_kafka_assignment(@native_kafka.inner, ptr)
  if response != 0
    raise Rdkafka::RdkafkaError.new(response)
  end

  tpl = ptr.read_pointer

  if !tpl.null?
    begin
      Rdkafka::Consumer::TopicPartitionList.from_native_tpl(tpl)
    ensure
      Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy tpl
    end
  end
ensure
  ptr.free unless ptr.nil?
end

#close ⇒ nil

Close this consumer

Returns:

  • (nil) 


27
28
29
30
31
32
 
# File 'lib/rdkafka/consumer.rb', line 27

def close
  return if closed?
  ObjectSpace.undefine_finalizer(self)
  Rdkafka::Bindings.rd_kafka_consumer_close(@native_kafka.inner)
  @native_kafka.close
end

#closed? ⇒ Boolean

Whether this consumer has closed

Returns:

  • (Boolean) 


35
36
37
 
# File 'lib/rdkafka/consumer.rb', line 35

def closed?
  @native_kafka.closed?
end

#cluster_id ⇒ String?

Returns the ClusterId as reported in broker metadata.

Returns:

  • (String, nil) 


307
308
309
310
 
# File 'lib/rdkafka/consumer.rb', line 307

def cluster_id
  closed_consumer_check(__method__)
  Rdkafka::Bindings.rd_kafka_clusterid(@native_kafka.inner)
end

#commit(list = nil, async = false) ⇒ nil

Manually commit the current offsets of this consumer.

To use this set enable.auto.committo false to disable automatic triggering of commits.

If enable.auto.offset.store is set to true the offset of the last consumed message for every partition is used. If set to false you can use #store_offset to indicate when a message has been fully processed.

Parameters:

  • list (TopicPartitionList, nil) (defaults to: nil) —

    The topic with partitions to commit

  • async (Boolean) (defaults to: false) —

    Whether to commit async or wait for the commit to finish

Returns:

  • (nil)

Raises:



403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
 
# File 'lib/rdkafka/consumer.rb', line 403

def commit(list=nil, async=false)
  closed_consumer_check(__method__)

  if !list.nil? && !list.is_a?(TopicPartitionList)
    raise TypeError.new("list has to be nil or a TopicPartitionList")
  end

  tpl = list ? list.to_native_tpl : nil

  begin
    response = Rdkafka::Bindings.rd_kafka_commit(@native_kafka.inner, tpl, async)
    if response != 0
      raise Rdkafka::RdkafkaError.new(response)
    end
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl) if tpl
  end
end

#committed(list = nil, timeout_ms = 1200) ⇒ TopicPartitionList

Return the current committed offset per partition for this consumer group. The offset field of each requested partition will either be set to stored offset or to -1001 in case there was no stored offset for that partition.

Parameters:

  • list (TopicPartitionList, nil) (defaults to: nil) —

    The topic with partitions to get the offsets for or nil to use the current subscription.

  • timeout_ms (Integer) (defaults to: 1200) —

    The timeout for fetching this information.

Returns:

Raises:

  • (RdkafkaError) —

    When getting the committed positions fails.



217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
 
# File 'lib/rdkafka/consumer.rb', line 217

def committed(list=nil, timeout_ms=1200)
  closed_consumer_check(__method__)

  if list.nil?
    list = assignment
  elsif !list.is_a?(TopicPartitionList)
    raise TypeError.new("list has to be nil or a TopicPartitionList")
  end

  tpl = list.to_native_tpl

  begin
    response = Rdkafka::Bindings.rd_kafka_committed(@native_kafka.inner, tpl, timeout_ms)
    if response != 0
      raise Rdkafka::RdkafkaError.new(response)
    end
    TopicPartitionList.from_native_tpl(tpl)
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl)
  end
end

#each {|message| ... } ⇒ nil

Poll for new messages and yield for each received one. Iteration will end when the consumer is closed.

If enable.partition.eof is turned on in the config this will raise an error when an eof is reached, so you probably want to disable that when using this method of iteration.

Yield Parameters:

  • message (Message) —

    Received message

Returns:

  • (nil)

Raises:



464
465
466
467
468
469
470
471
472
473
474
475
476
477
 
# File 'lib/rdkafka/consumer.rb', line 464

def each
  loop do
    message = poll(250)
    if message
      yield(message)
    else
      if closed?
        break
      else
        next
      end
    end
  end
end

#each_batch(max_items: 100, bytes_threshold: Float::INFINITY, timeout_ms: 250, yield_on_error: false) {|messages, pending_exception| ... } ⇒ nil

Poll for new messages and yield them in batches that may contain messages from more than one partition.

Rather than yield each message immediately as soon as it is received, each_batch will attempt to wait for as long as timeout_ms in order to create a batch of up to but no more than max_items in size.

Said differently, if more than max_items are available within timeout_ms, then each_batch will yield early with max_items in the array, but if timeout_ms passes by with fewer messages arriving, it will yield an array of fewer messages, quite possibly zero.

In order to prevent wrongly auto committing many messages at once across possibly many partitions, callers must explicitly indicate which messages have been successfully processed as some consumed messages may not have been yielded yet. To do this, the caller should set enable.auto.offset.store to false and pass processed messages to #store_offset. It is also possible, though more complex, to set 'enable.auto.commit' to false and then pass a manually assembled TopicPartitionList to #commit.

As with each, iteration will end when the consumer is closed.

Exception behavior is more complicated than with each, in that if :yield_on_error is true, and an exception is raised during the poll, and messages have already been received, they will be yielded to the caller before the exception is allowed to propagate.

If you are setting either auto.commit or auto.offset.store to false in the consumer configuration, then you should let yield_on_error keep its default value of false because you are guaranteed to see these messages again. However, if both auto.commit and auto.offset.store are set to true, you should set yield_on_error to true so you can process messages that you may or may not see again.

which will be propagated after processing of the partial batch is complete.

Parameters:

  • max_items (Integer) (defaults to: 100) —

    Maximum size of the yielded array of messages

  • bytes_threshold (Integer) (defaults to: Float::INFINITY) —

    Threshold number of total message bytes in the yielded array of messages

  • timeout_ms (Integer) (defaults to: 250) —

    max time to wait for up to max_items

Yields:

  • (messages, pending_exception)

Yield Parameters:

  • messages (Array) —

    An array of received Message

  • pending_exception (Exception) —

    normally nil, or an exception

Returns:

  • (nil)

Raises:



528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
 
# File 'lib/rdkafka/consumer.rb', line 528

def each_batch(max_items: 100, bytes_threshold: Float::INFINITY, timeout_ms: 250, yield_on_error: false, &block)
  closed_consumer_check(__method__)
  slice = []
  bytes = 0
  end_time = monotonic_now + timeout_ms / 1000.0
  loop do
    break if closed?
    max_wait = end_time - monotonic_now
    max_wait_ms = if max_wait <= 0
                    0  # should not block, but may retrieve a message
                  else
                    (max_wait * 1000).floor
                  end
    message = nil
    begin
      message = poll max_wait_ms
    rescue Rdkafka::RdkafkaError => error
      raise unless yield_on_error
      raise if slice.empty?
      yield slice.dup, error
      raise
    end
    if message
      slice << message
      bytes += message.payload.bytesize if message.payload
    end
    if slice.size == max_items || bytes >= bytes_threshold || monotonic_now >= end_time - 0.001
      yield slice.dup, nil
      slice.clear
      bytes = 0
      end_time = monotonic_now + timeout_ms / 1000.0
    end
  end
end

#finalizer ⇒ Object 



21
22
23
 
# File 'lib/rdkafka/consumer.rb', line 21

def finalizer
  ->(_) { close }
end

#lag(topic_partition_list, watermark_timeout_ms = 100) ⇒ Hash<String, Hash<Integer, Integer>>

Calculate the consumer lag per partition for the provided topic partition list. You can get a suitable list by calling #committed or position (TODO). It is also possible to create one yourself, in this case you have to provide a list that already contains all the partitions you need the lag for.

Parameters:

  • topic_partition_list (TopicPartitionList) —

    The list to calculate lag for.

  • watermark_timeout_ms (Integer) (defaults to: 100) —

    The timeout for each query watermark call.

Returns:

  • (Hash<String, Hash<Integer, Integer>>) —

    A hash containing all topics with the lag per partition

Raises:



283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
 
# File 'lib/rdkafka/consumer.rb', line 283

def lag(topic_partition_list, watermark_timeout_ms=100)
  out = {}

  topic_partition_list.to_h.each do |topic, partitions|
    # Query high watermarks for this topic's partitions
    # and compare to the offset in the list.
    topic_out = {}
    partitions.each do |p|
      next if p.offset.nil?
      low, high = query_watermark_offsets(
        topic,
        p.partition,
        watermark_timeout_ms
      )
      topic_out[p.partition] = high - p.offset
    end
    out[topic] = topic_out
  end
  out
end

#member_id ⇒ String?

Returns this client's broker-assigned group member id

This currently requires the high-level KafkaConsumer

Returns:

  • (String, nil) 


317
318
319
320
 
# File 'lib/rdkafka/consumer.rb', line 317

def member_id
  closed_consumer_check(__method__)
  Rdkafka::Bindings.rd_kafka_memberid(@native_kafka.inner)
end

#pause(list) ⇒ nil

Pause producing or consumption for the provided list of partitions

Parameters:

Returns:

  • (nil)

Raises:



86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
 
# File 'lib/rdkafka/consumer.rb', line 86

def pause(list)
  closed_consumer_check(__method__)

  unless list.is_a?(TopicPartitionList)
    raise TypeError.new("list has to be a TopicPartitionList")
  end

  tpl = list.to_native_tpl

  begin
    response = Rdkafka::Bindings.rd_kafka_pause_partitions(@native_kafka.inner, tpl)

    if response != 0
      list = TopicPartitionList.from_native_tpl(tpl)
      raise Rdkafka::RdkafkaTopicPartitionListError.new(response, list, "Error pausing '#{list.to_h}'")
    end
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl)
  end
end

#poll(timeout_ms) ⇒ Message?

Poll for the next message on one of the subscribed topics

Parameters:

  • timeout_ms (Integer) —

    Timeout of this poll

Returns:

  • (Message, nil) —

    A message or nil if there was no new message within the timeout

Raises:



429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
 
# File 'lib/rdkafka/consumer.rb', line 429

def poll(timeout_ms)
  closed_consumer_check(__method__)

  message_ptr = Rdkafka::Bindings.rd_kafka_consumer_poll(@native_kafka.inner, timeout_ms)
  if message_ptr.null?
    nil
  else
    # Create struct wrapper
    native_message = Rdkafka::Bindings::Message.new(message_ptr)
    # Raise error if needed
    if native_message[:err] != 0
      raise Rdkafka::RdkafkaError.new(native_message[:err])
    end
    # Create a message to pass out
    Rdkafka::Consumer::Message.new(native_message)
  end
ensure
  # Clean up rdkafka message if there is one
  if message_ptr && !message_ptr.null?
    Rdkafka::Bindings.rd_kafka_message_destroy(message_ptr)
  end
end

#query_watermark_offsets(topic, partition, timeout_ms = 200) ⇒ Integer

Query broker for low (oldest/beginning) and high (newest/end) offsets for a partition.

Parameters:

  • topic (String) —

    The topic to query

  • partition (Integer) —

    The partition to query

  • timeout_ms (Integer) (defaults to: 200) —

    The timeout for querying the broker

Returns:

  • (Integer) —

    The low and high watermark

Raises:



248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
 
# File 'lib/rdkafka/consumer.rb', line 248

def query_watermark_offsets(topic, partition, timeout_ms=200)
  closed_consumer_check(__method__)

  low = FFI::MemoryPointer.new(:int64, 1)
  high = FFI::MemoryPointer.new(:int64, 1)

  response = Rdkafka::Bindings.rd_kafka_query_watermark_offsets(
    @native_kafka.inner,
    topic,
    partition,
    low,
    high,
    timeout_ms,
  )
  if response != 0
    raise Rdkafka::RdkafkaError.new(response, "Error querying watermark offsets for partition #{partition} of #{topic}")
  end

  return low.read_array_of_int64(1).first, high.read_array_of_int64(1).first
ensure
  low.free   unless low.nil?
  high.free  unless high.nil?
end

#resume(list) ⇒ nil

Resume producing consumption for the provided list of partitions

Parameters:

Returns:

  • (nil)

Raises:



114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
 
# File 'lib/rdkafka/consumer.rb', line 114

def resume(list)
  closed_consumer_check(__method__)

  unless list.is_a?(TopicPartitionList)
    raise TypeError.new("list has to be a TopicPartitionList")
  end

  tpl = list.to_native_tpl

  begin
    response = Rdkafka::Bindings.rd_kafka_resume_partitions(@native_kafka.inner, tpl)
    if response != 0
      raise Rdkafka::RdkafkaError.new(response, "Error resume '#{list.to_h}'")
    end
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl)
  end
end

#seek(message) ⇒ nil

Seek to a particular message. The next poll on the topic/partition will return the message at the given offset.

Parameters:

Returns:

  • (nil)

Raises:



363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
 
# File 'lib/rdkafka/consumer.rb', line 363

def seek(message)
  closed_consumer_check(__method__)

  # rd_kafka_offset_store is one of the few calls that does not support
  # a string as the topic, so create a native topic for it.
  native_topic = Rdkafka::Bindings.rd_kafka_topic_new(
    @native_kafka.inner,
    message.topic,
    nil
  )
  response = Rdkafka::Bindings.rd_kafka_seek(
    native_topic,
    message.partition,
    message.offset,
    0 # timeout
  )
  if response != 0
    raise Rdkafka::RdkafkaError.new(response)
  end
ensure
  if native_topic && !native_topic.null?
    Rdkafka::Bindings.rd_kafka_topic_destroy(native_topic)
  end
end

#store_offset(message) ⇒ nil

Store offset of a message to be used in the next commit of this consumer

When using this enable.auto.offset.store should be set to false in the config.

Parameters:

Returns:

  • (nil)

Raises:



331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
 
# File 'lib/rdkafka/consumer.rb', line 331

def store_offset(message)
  closed_consumer_check(__method__)

  # rd_kafka_offset_store is one of the few calls that does not support
  # a string as the topic, so create a native topic for it.
  native_topic = Rdkafka::Bindings.rd_kafka_topic_new(
    @native_kafka.inner,
    message.topic,
    nil
  )
  response = Rdkafka::Bindings.rd_kafka_offset_store(
    native_topic,
    message.partition,
    message.offset
  )
  if response != 0
    raise Rdkafka::RdkafkaError.new(response)
  end
ensure
  if native_topic && !native_topic.null?
    Rdkafka::Bindings.rd_kafka_topic_destroy(native_topic)
  end
end

#subscribe(*topics) ⇒ nil

Subscribe to one or more topics letting Kafka handle partition assignments.

Parameters:

  • topics (Array<String>) —

    One or more topic names

Returns:

  • (nil)

Raises:



46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
 
# File 'lib/rdkafka/consumer.rb', line 46

def subscribe(*topics)
  closed_consumer_check(__method__)

  # Create topic partition list with topics and no partition set
  tpl = Rdkafka::Bindings.rd_kafka_topic_partition_list_new(topics.length)

  topics.each do |topic|
    Rdkafka::Bindings.rd_kafka_topic_partition_list_add(tpl, topic, -1)
  end

  # Subscribe to topic partition list and check this was successful
  response = Rdkafka::Bindings.rd_kafka_subscribe(@native_kafka.inner, tpl)
  if response != 0
    raise Rdkafka::RdkafkaError.new(response, "Error subscribing to '#{topics.join(', ')}'")
  end
ensure
  Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(tpl) unless tpl.nil?
end

#subscription ⇒ TopicPartitionList

Return the current subscription to topics and partitions

Returns:

Raises:



138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
 
# File 'lib/rdkafka/consumer.rb', line 138

def subscription
  closed_consumer_check(__method__)

  ptr = FFI::MemoryPointer.new(:pointer)
  response = Rdkafka::Bindings.rd_kafka_subscription(@native_kafka.inner, ptr)

  if response != 0
    raise Rdkafka::RdkafkaError.new(response)
  end

  native = ptr.read_pointer

  begin
    Rdkafka::Consumer::TopicPartitionList.from_native_tpl(native)
  ensure
    Rdkafka::Bindings.rd_kafka_topic_partition_list_destroy(native)
  end
end

#unsubscribe ⇒ nil

Unsubscribe from all subscribed topics.

Returns:

  • (nil)

Raises:



70
71
72
73
74
75
76
77
 
# File 'lib/rdkafka/consumer.rb', line 70

def unsubscribe
  closed_consumer_check(__method__)

  response = Rdkafka::Bindings.rd_kafka_unsubscribe(@native_kafka.inner)
  if response != 0
    raise Rdkafka::RdkafkaError.new(response)
  end
end